Voice activity detection system and sound feature extraction circuit therefor

Abstract

The present invention is a voice activity detection system and a sound feature extraction circuit thereof. The sound feature extraction circuit includes a plurality of band-pass filters corresponding to a plurality of channels, respectively having different band-pass frequency ranges, for switching to receive an amplified signal to generate corresponding filtered signals, an operational amplifier shared by the plurality of channels, and a rectifier switched to receive the filtered signals to generate a rectified signal. The amplified signal is transmitted to the plurality of band-pass filters at different phases in a time-division multiplexing manner, and the filtered signals are transmitted to the rectifier at different phases in a time-division multiplexing manner.

Description

TECHNICAL FIELD

[0001] The present application relates to a voice activity detection system, and more particularly to a sound feature extraction circuit. BACKGROUND

[0002] Voice activity detection (VAD) can be used to detect or recognize human speech. Voice activity detection can trigger voice-based applications, such as Apple's virtual assistant Siri. Voice activity detection is a front-end device that is usually always-on and a low-power system.

[0003] Conventional voice activity detection uses a digital architecture to implement, thus occupying a large circuit area and consuming a large power supply. In addition, conventional voice activity detection is limited by process-voltage-temperature variation (PVT variation).

[0004] Therefore, there is an urgent need to propose a novel voice activity detection system to overcome the shortcomings of conventional voice activity detection systems. SUMMARY

[0005] In view of the above, one of the purposes of the embodiments of the present application is to propose a voice activity detection system, in which the sound feature extraction circuit can effectively reduce power consumption and has high performance.

[0006] According to the embodiments of the present application, the sound feature extraction circuit includes a plurality of band-pass filters and a rectifier. The band-pass filters correspond to a plurality of channels, respectively having different band-pass frequency ranges, and are switched to receive an amplified signal to generate a corresponding filtered signal. An operational amplifier is shared by the plurality of channels. The rectifier is switched to receive the filtered signal to generate a rectified signal. The amplified signal is transmitted to the plurality of band-pass filters at different phases in a time-division demultiplexing manner, and the filtered signal is transmitted to the rectifier at different phases in a time-division multiplexing manner.

[0007] Preferably, further comprising: a plurality of phase switches, the amplified signal being transmitted to the plurality of band-pass filters at different phases in a time-division demultiplexing manner by the plurality of phase switches.

[0008] Preferably, further comprising: a plurality of phase switches, the filtered signal being transmitted to the rectifier at different phases in a time-division multiplexing manner by the plurality of phase switches.

[0009] Preferably, each band-pass filter includes: a first stabilizing capacitor, a first end of the first stabilizing capacitor being switched to connect to a negative input node of the operational amplifier by a phase switch, the phase switch being turned on at a corresponding phase, a second end of the first stabilizing capacitor being connected to ground; wherein each phase is divided into a first period and a second period.

[0010] Preferably, each band-pass filter further comprises a second stabilization capacitor having a first end switchably connected to an output node of the operational amplifier by another phase switch, the another phase switch being conductive with respect to a corresponding phase, and a second end connected to ground.

[0011] Preferably, each band-pass filter further comprises a first charging capacitor having a first end switchably receiving the amplified signal by a first period switch, the first period switch being conductive with respect to a first period, and a second end connected to ground; a first filter capacitor having a first end switchably connected to the first end of the first charging capacitor by a second period switch, the second period switch being conductive with respect to a second period, and a second end switchably connected to a negative input node of the operational amplifier by the phase switch; a second filter capacitor having a first end connected to the first end of the first filter capacitor, and a second end switchably connected to an output node of the operational amplifier by another phase switch, the another phase switch being conductive with respect to a corresponding phase; and a second charging capacitor having a first end switchably connected to the second end of the first filter capacitor by another first period switch, and switchably connected to the second end of the second filter capacitor by another second period switch, and a second end connected to ground.

[0012] According to an embodiment of the present application, a voice activity detection system comprises an amplifier receiving a sound signal to generate an amplified signal; a sound feature extraction circuit receiving the amplified signal to generate a feature signal representing a feature extracted from the amplified signal; and a classifier identifying the feature signal as sound or noise; wherein the sound feature extraction circuit comprises a single operational amplifier shared by a plurality of channels.

[0013] Preferably, the sound feature extraction circuit comprises a plurality of band-pass filters having different band-pass frequency ranges with respect to the plurality of channels, respectively, switchably receiving the amplified signal to generate corresponding filtered signals; and a rectifier switchably receiving the filtered signals to generate a rectified signal; wherein the amplified signal is transmitted to the plurality of band-pass filters in different phases in a time-division multiplexing manner, and the filtered signals are transmitted to the rectifier in different phases in a time-division multiplexing manner.

[0014] Preferably, each band-pass filter comprises the operational amplifier shared by the plurality of channels; and a first stabilization capacitor having a first end switchably connected to a negative input node of the operational amplifier by a phase switch, the phase switch being conductive with respect to a corresponding phase, and a second end connected to ground; wherein each phase is divided into a first period and a second period.

[0015] Preferably, each band-pass filter further comprises a second stabilization capacitor having a first end connected to the output node of the operational amplifier by another phase switch which is turned on with respect to the corresponding phase, and a second end connected to ground. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A block diagram of a voice activity detection system according to an embodiment of the present application is shown.

[0017] Figure 2 A detailed block diagram of a sound feature extraction circuit according to an embodiment of the present application is shown. Figure 1

[0018] Figure 3A A circuit diagram of a band-pass filter corresponding to a channel according to an embodiment of the present application is shown. Figure 2

[0019] A timing diagram showing the time period of a corresponding phase according to an embodiment of the present application is shown. Figure 3B

[0020] A circuit diagram of a band-pass filter corresponding to a three-channel according to an embodiment of the present application is shown. Figure 4A

[0021] A timing diagram showing the time period of a corresponding phase according to an embodiment of the present application is shown. Figure 4B

[0022] An equivalent circuit of a band-pass filter at a first period Φ1 of a first phase P1 according to an embodiment of the present application is shown. Figure 5A

[0023] An equivalent circuit of a band-pass filter at a second period Φ2 of a first phase P1 according to an embodiment of the present application is shown. Figure 5B

[0024] An equivalent circuit of a band-pass filter at a first period Φ3 of a second phase P2 according to an embodiment of the present application is shown. Figure 5C

[0025] A table showing the state of a band-pass filter at a corresponding phase and period according to an embodiment of the present application is shown. Figure 5D

[0026] A circuit diagram of a rectifier according to an embodiment of the present application is shown. Figure 6 Figure 2 A circuit diagram of a low-pass filter corresponding to a channel according to an embodiment of the present application is shown.

[0027] Figure 7

MAIN ELEMENT SYMBOL EXPLANATION

[0028] 100: voice activity detection system 10: microphone

[0029] 11: amplifier 12: sound feature extraction circuit

[0030] 11: amplifier 12: sound feature extraction circuit ​​

[0031] 121: band-pass filter 1211: operational amplifier

[0032] 122: rectifier 1221: operational amplifier

[0033] 123: low-pass filter 1231: operational amplifier

[0034] 13: classifier 14: buffer

[0035] P1-P3: phase switch / phase Φ1: first period / first duration switch

[0036] Φ2: second period / second duration switch CR1: first charge capacitor

[0037] CR2: second charge capacitor CC1: first filter capacitor / filter capacitor

[0038] CC2: second filter capacitor CL1: first stabilization capacitor

[0039] CL2: second stabilization capacitor VCM: common mode voltage

[0040] ΦC: first switch / switching signal ΦC: second switch / inverted switching signal DETAILED DESCRIPTION

[0041] Figure 1 A block diagram of a voice activity detection (VAD) system 100 according to an embodiment of the present application is shown. In this embodiment, the voice activity detection system 100 can be an analog system that operates on analog signals.

[0042] The voice activity detection system 100 according to this embodiment can include an amplifier 11 that receives a sound signal from a transducer (e.g., a microphone 10) that converts a sound signal into an amplified signal. In this embodiment, the amplifier 11 can be a low-noise amplifier (LNA) that amplifies a low-power signal (e.g., a sound signal) without degrading the signal-to-noise ratio (SNR).

[0043] The voice activity detection system 100 according to this embodiment can include an acoustic feature extraction (AFE) circuit 12 that receives the amplified signal and generates a feature signal that represents features extracted from the amplified signal.

[0044] The voice activity detection system 100 of this embodiment may include a classifier 13 for identifying feature signals as voice or analog. In one embodiment, the classifier 13 may include (analog) neural network circuitry.

[0045] In one embodiment, the voice activity detection system 100 may further include a buffer 14 (e.g., a unit-gain buffer) located between the amplifier 11 and the sound feature acquisition circuit 12 to provide impedance switching from the amplifier 11 to the sound feature acquisition circuit 12, so that the amplified signal is not affected by the sound feature acquisition circuit 12 (i.e., the load).

[0046] Figure 2 show Figure 1 A detailed block diagram of the sound feature acquisition circuit 12 is provided. In this embodiment, the sound feature acquisition circuit 12 may include multiple band-pass filters (BPFs) 121, corresponding to multiple channels (three channels in this example), each with a different bandpass frequency range (e.g., 95–195 Hz, 150–310 Hz, and 250–500 Hz), thus generating corresponding filtered signals. The band-pass filters 121 are switched to receive amplified signals (from amplifier 11). According to one feature of this embodiment, the amplified signals are transmitted to the band-pass filters 121 in a time-division demultiplexing manner at different phases (e.g., transmitted via phase switches P1–P3 in phases 1 to 3, respectively).

[0047] The sound feature acquisition circuit 12 in this embodiment may include a rectifier 122, which is switched to receive a filtered signal. At different phases, the filtered signal is transmitted to the rectifier 122 in a time-division multiplex manner (for example, transmitted separately via phase switches P1 to P3), thereby generating a rectified signal. In this way, only a single rectifier 122 is needed for all channels.

[0048] The sound feature extraction circuit 12 in this embodiment may include multiple low-pass filters (LPFs) 123, corresponding to multiple channels, having the same low-pass frequency range (e.g., a cutoff frequency of 30Hz), thus generating corresponding feature signals (e.g., Figure 2 (The first to third characteristic signals are illustrated). The low-pass filter 123 is switched to receive the rectified signal (from the rectifier 122). At different phases, the rectified signal is transmitted to the low-pass filter 123 in a time-division multiplexing manner (for example, transmitted via phase switches P1 to P3 in phases 1 to 3 respectively).

[0049] Figure 3A Display Figure 2 A circuit diagram of the band-pass filter 121 corresponding to a channel (e.g. channel 1), Figure 3B A timing diagram showing the time duration of each phase. In this embodiment, each phase is divided into a first period (e.g. Φ1) and a second period (e.g. Φ2). For example, the first period Φ1 is located at the first half of the phase signal P1, and the second period Φ2 follows the first period Φ1 until the beginning of the next phase signal P1.

[0050] In this embodiment, the band-pass filter 121 can comprise a switched-capacitor (SC) circuit. The band-pass filter 121 can comprise an operational amplifier 1211 having a negative input node, a positive input node (connected to ground), and an output node. According to one of the features of the present embodiment, multiple channels can share a single operational amplifier 1211 in a time-division multiplexing manner, thus greatly reducing power consumption and circuit area.

[0051] The band-pass filter 121 can comprise a first charging capacitor CR1 having a first end switched to receive the amplified signal via a first period switch Φ1 (which is turned on during the first period) and a second end connected to ground. The band-pass filter 121 can comprise a first filter capacitor CC1 having a first end switched to be connected to the first end of the first charging capacitor CR1 via a second period switch Φ2 (which is turned on during the second period) and a second end switched to be connected to the negative input node of the operational amplifier 1211 via a phase switch P1 (which is turned on during the corresponding phase). The band-pass filter 121 can comprise a second filter capacitor CC2 having a first end connected to the first end of the first filter capacitor CC1 and a second end switched to be connected to the output node of the operational amplifier 1211 via another phase switch P1. The band-pass filter 121 can comprise a second charging capacitor CR2 having a first end switched to be connected to the second end of the first filter capacitor CC1 via another first period switch Φ1 and to the second end of the second filter capacitor CC2 via another second period switch Φ2, and a second end connected to ground.

[0052] According to another feature of the present embodiment, the bandpass filter 121 can include a first stabilization capacitor CL1 having a first end connected to the negative input node of the operational amplifier 1211 via the phase switch P1 and a second end connected to ground. The first stabilization capacitor CL1 can solve the floating voltage problem of the connection node between the first filter capacitor CC1 and the first stabilization capacitor CL1 in the hold state. The bandpass filter 121 can further include a second stabilization capacitor CL2 having a first end connected to the output node of the operational amplifier 1211 via the other phase switch P1 and a second end connected to ground. The second stabilization capacitor CL2 can be used to stabilize the voltage of the output node of the operational amplifier 1211 in the charge state.

[0053] Figure 4A a circuit diagram showing the bandpass filter 121 corresponding to three channels, Figure 4B a timing diagram showing the time period of the corresponding phase. It is noted that all channels can share a single operational amplifier 1211 in a time-division multiplexing manner.

[0054] Figure 5A an equivalent circuit of the bandpass filter 121 in the first period Φ1 of the first phase P1, in which the bandpass filter 121 of channel 1 is in the charge state, the bandpass filter 121 of channel 2 is in the hold state, and the bandpass filter 121 of channel 3 is in the hold state. Figure 5B an equivalent circuit of the bandpass filter 121 in the second period Φ2 of the first phase P1, in which the bandpass filter 121 of channel 1 is in the filter state, the bandpass filter 121 of channel 2 is in the hold state, and the bandpass filter 121 of channel 3 is in the hold state. Figure 5C an equivalent circuit of the bandpass filter 121 in the first period Φ3 of the second phase P2, in which the bandpass filter 121 of channel 1 is in the rectification state, the bandpass filter 121 of channel 2 is in the charge state, and the bandpass filter 121 of channel 3 is in the hold state. Figure 5D a table showing the states of the bandpass filter 121 in the corresponding phase and period. It is noted that the operational amplifier 1211 is used in the charge state and the filter state, and the rectifier 122 is used in the rectification state.

[0055] Figure 6 a circuit diagram showing Figure 2The circuit diagram of the rectifier 122 is shown in FIG. 3. In this embodiment, the rectifier 122 can include an operational amplifier 1221 having a positive input node receiving the filtered signal and a negative input node receiving a common-mode voltage VCM, thus generating a switching signal ΦC. The filtered signal is switched to be transmitted through a first switch ΦC controlled by the switching signal ΦC, and the common-mode voltage VCM is switched to be transmitted through a second switch ΦC controlled by an inverted switching signal ΦC. In this way, one of the filtered signal and the common-mode voltage is transmitted to become a rectified signal, thus forming a half-wave rectifier. In another alternative embodiment, the rectifier 122 can be implemented as a full-wave rectifier.

[0056] Figure 7 The circuit diagram of the low-pass filter 123 corresponding to the channel is shown in FIG. 4. In this embodiment, the low-pass filter 123 can include a switched-capacitor circuit. The low-pass filter 123 can include an operational amplifier 1231 having a negative input node, a positive input node (connected to ground), and an output node. The operational amplifier 1231 can be omitted if no buffering or driving is needed. The low-pass filter 123 can include a first charging capacitor CR1 having a first end switched to receive the rectified signal through a first period switch Φ1 (which is turned on in a first period) and a second end connected to ground. The low-pass filter 123 can include a filter capacitor CC1 having a first end switched to be connected to the first end of the first charging capacitor CR1 through a second period switch Φ2 (which is turned on in a second period) and a second end connected to the output node of the operational amplifier 1231. The low-pass filter 123 can include a second charging capacitor CR2 having a first end switched to be connected to the negative input node of the operational amplifier 1231 through another first period switch Φ1 and to the second end of the filter capacitor CC1 through another second period switch Φ2, and a second end connected to ground.

[0057] The above merely provides the preferred embodiments of the present application, but is not intended to limit the present application in any manner. Although the present application has been disclosed based on the preferred embodiments, it is not intended to limit the present application, and any person skilled in the art can make some changes or modifications to the above disclosed technical contents to obtain equivalent embodiments with equivalent changes, as long as the changes or modifications do not depart from the technical solutions of the present application. Any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present application shall still fall within the scope of the technical solutions of the present application.

Claims

1. A sound feature acquisition circuit, characterized in that, Include: Multiple bandpass filters, corresponding to multiple channels, each have different bandpass frequency ranges. These filters switch the received amplified signal to generate a corresponding filtered signal. Each bandpass filter includes an operational amplifier shared by the multiple channels. The rectifier is switched to receive the filtered signal in order to generate a rectified signal; The amplified signal is transmitted to the multiple bandpass filters at different phases using a time-division multiplexing method, and the filtered signal is transmitted to the rectifier at different phases using a time-division multiplexing method. Each bandpass filter includes a first stabilizing capacitor, the first end of which is switched to the negative input node of the operational amplifier by a phase switch, the phase switch being on for the corresponding phase, and the second end of the first stabilizing capacitor being connected to ground.

2. The sound feature acquisition circuit according to claim 1, characterized in that, It also includes: Multiple phase switches are used to transmit the amplified signal to the multiple bandpass filters at different phases in a time-division multiplexing manner.

3. The sound feature acquisition circuit according to claim 1, characterized in that, It also includes: Multiple phase switches are used to transmit the filtered signal to the rectifier in a time-division multiplexing manner at different phases.

4. The sound feature acquisition circuit according to claim 1, characterized in that, Each phase is divided into a first period and a second period.

5. The sound feature acquisition circuit according to claim 4, characterized in that, Each bandpass filter further includes: The second stabilizing capacitor has its first terminal connected to the output node of the operational amplifier via another phase switch, which is turned on for the corresponding phase, and its second terminal is connected to ground.

6. The sound feature acquisition circuit according to claim 4, characterized in that, Each bandpass filter further includes: A first charging capacitor, the first terminal of which is switched to receive the amplified signal by a first period switch, the first period switch being on during the first period, and the second terminal of the first charging capacitor being connected to ground. A first filter capacitor, the first terminal of which is switched to the first terminal of the first charging capacitor by means of a second period switch, the second period switch being on during the second period, and the second terminal of the first filter capacitor being switched to the negative input node of the operational amplifier by means of the phase switch. The second filter capacitor has its first terminal connected to the first terminal of the first filter capacitor, and its second terminal is switched to the output node of the operational amplifier by means of another phase switch, wherein the other phase switch is on for the corresponding phase; and The second charging capacitor has its first terminal switched to the second terminal of the first filter capacitor by another first period switch and its second terminal switched to the second filter capacitor by another second period switch. The second terminal of the second charging capacitor is connected to ground.

7. A voice activity detection system, characterized in that, Include: An amplifier receives sound signals and uses them to generate amplified signals; A sound feature extraction circuit receives the amplified signal and generates a feature signal representing the features extracted from the amplified signal; and A classifier is used to identify whether the characteristic signal is sound or analog. The sound feature acquisition circuit includes a single operational amplifier, which is shared by multiple channels; The sound feature extraction circuit includes: Multiple bandpass filters, corresponding to the multiple channels, each have different bandpass frequency ranges, and the received amplified signal is switched to generate the corresponding filtered signal; and The rectifier is switched to receive the filtered signal in order to generate a rectified signal; The amplified signal is transmitted to the multiple bandpass filters at different phases in a time-division multiplexing manner, and the filtered signal is transmitted to the rectifier at different phases in a time-division multiplexing manner. Each bandpass filter contains: The operational amplifier is shared by these multiple channels; and A first stabilizing capacitor, the first terminal of which is switched to the negative input node of the operational amplifier by means of a phase switch, the phase switch being on for the corresponding phase, and the second terminal of the first stabilizing capacitor being connected to ground.

8. The voice activity detection system according to claim 7, characterized in that, Each phase is divided into a first period and a second period.

9. The voice activity detection system according to claim 8, characterized in that, Each bandpass filter further includes: The second stabilizing capacitor has its first terminal connected to the output node of the operational amplifier via another phase switch, which is turned on for the corresponding phase, and its second terminal is connected to ground.

Images